DNMT3A mutation is associated with increased age and adverse outcome in adult T-cell acute lymphoblastic leukemia

Abstract

The prognostic implications of DNMT3A genotype in T-cell acute lymphoblastic leukemia are incompletely understood. We performed comprehensive genetic and clinico-biological analyses of T-cell acute lymphoblastic leukemia patients with DNMT3A mutations treated during the GRAALL-2003 and -2005 studies. Eighteen of 198 cases (9.1%) had DNMT3A alterations. Two patients also had DNMT3A mutations in non-leukemic cell DNA, providing the first potential evidence of age-related clonal hematopoiesis in T-cell acute lymphoblastic leukemia. DNMT3A mutation was associated with older age (median 43.9 years vs. 29.4 years, P<0.001), immature T-cell receptor genotype (53.3% vs. 24.4%, P=0.016) and lower remission rates (72.2% mutated vs. 94.4% non-mutated, P=0.006). DNMT3A alterations were significantly associated with worse clinical outcome, with higher cumulative incidence of relapse (HR 2.33, 95% CI: 1.05-5.16, P=0.037) and markedly poorer event-free survival (HR 3.22, 95% CI: 1.81-5.72, P<0.001) and overall survival (HR 2.91, 95% CI: 1.56-5.43, P=0.001). Adjusting for age as a covariate, or restricting the analysis to patients over 40 years, who account for almost 90% of DNMT3A-mutated cases, did not modify these observations. In multivariate analysis using the risk factors that were used to stratify treatment during the GRAALL studies, DNMT3A mutation was significantly associated with shorter event-free survival (HR 2.33, 95% CI: 1.06 – 4.04, P=0.02). Altogether, these results identify DNMT3A genotype as a predictor of aggressive T-cell acute lymphoblastic leukemia biology. The GRAALL-2003 and -2005 studies were registered at http://www.ClinicalTrials.gov as #NCT00222027 and #NCT00327678, respectively.

Introduction

Mutations in the DNA methyltransferase 3 alpha gene (DNMT3A) have been reported in a range of hematologic malignancies, most frequently in myeloid neoplasia, including acute myeloid leukemia (AML),51 myelodysplastic syndromes,6 myeloproliferative neoplasms7 and myeloproliferative neoplasm/myelodysplastic overlap syndromes.98 DNMT3A alterations in lymphoid malignancies are less common, and reports to date are confined to T-lineage disease.1610 In all cases, DNMT3A mutations increase in frequency with age, and are extremely rare in children and adolescents.1917

Multiple studies have reported that DNMT3A alterations correlate with poor outcome in AML.2220421 In comparison, the prognostic influence of DNMT3A mutation in T-cell acute lymphoblastic leukemia (T-ALL) is poorly characterized. Patients with DNMT3A alterations were reported to have shorter survival in three moderately sized (55 to 93 patients) T-ALL cohorts.13119 DNMT3A status did not however independently predict prognosis in the only series for which multivariate analyses were documented, as survival effects were linked to increased rates of DNMT3A mutation in poor-risk, phenotypically immature disease.11 While that study did document a correlation between DNMT3A alteration and survival within the immature T-ALL subgroup, this finding was not corroborated in an independent cohort of early thymic precursor (ETP) ALL cases.12

The issue of whether DNMT3A mutation truly alters the biology of T-ALL is therefore only partially addressed by the currently available evidence. In particular, it is unclear whether the associated poor survival simply reflects the prosaic fact that patients with DNMT3A alterations are older,1211 and therefore do not tolerate intensive ALL treatment as well as their younger counterparts.

In order to address this question, we used next-generation sequencing (NGS) to evaluate the DNMT3A genotype of a large cohort of 198 adult T-ALL patients treated as part of the multinational GRAALL-2003 and -2005 studies. We found that DNMT3A mutation strongly correlated with disease relapse and shorter survival, and that these prognostic effects were independent of patients’ age. Furthermore, we report the presence of DNMT3A mutations in non-leukemic cells in a subset of patients, providing the first evidence of age-related clonal hematopoiesis in T-ALL.

Methods

Patients

Details of the GRAALL-2003 and -2005 studies are provided in the Online Supplementary Methods. Informed consent was obtained from all patients before inclusion into the trials. Both studies were conducted in accordance with the Declaration of Helsinki and approved by local and multicenter research ethical committees. The complete study protocols are detailed in the Online Data Supplement. Both trials were registered at http:/www.ClinicalTrials.gov (NCT00222027, NCT00327678). The criteria for inclusion in the current project were a diagnosis of TALL and the availability of diagnostic material for NGS analysis of DNMT3A genotype. Survival outcomes of the 198 patients (36 from GRAALL-2003 and 162 from GRAALL-2005) who fulfilled these criteria did not differ from those of the remaining 139 T-ALL patients of the study cohorts. As expected in retro spective studies, initial white blood cell count (WBC) was higher in the study cohort. However, no differences in allogeneic stem cell transplant rate, disease-free survival, event-free survival, or overall survival were found. A full comparison of the clinical features of each group is shown in Online Supplementary Table S1.

Next-generation sequencing

Nextera XT (Illumina) DNA Libraries were prepared according to the manufacturer’s instructions and sequenced using the Illumina MiSeq sequencing system. The custom NGS panel comprised genes coding for factors involved in molecular pathways known to be mutated in T-ALL, namely cytokine receptor and RAS signaling (NRAS, KRAS, JAK1, JAK3, STAT3, STAT5B, IL7R, BRAF, NF1, SH2B3, PTPN11), hematopoietic development (RUNX1, ETV6, GATA3, IKZF1, EP300), chemical modification of histones (SUZ12, EED, EZH2, KMT2A, KMT2D, SETD2) and DNA methylation (DNMT3A, IDH1, IDH2, TET2, TET3). This panel was originally inspired by the repertoire of genes found to be preferentially altered in pediatric ETP-ALL,23 and we have reported a subset of the results described in the current paper in a previous clinicobiological and genetic analysis of adult ETP-ALL.24 Sequencing reads were analyzed using in-house software (Polyweb, Institut Imagine, Paris, France), and additional in-house custom filtering criteria (comprising minimum read counts and variant allele frequencies, and reference to external reference databases) were applied to minimize false-positive rates. Primers used to confirm mutations by direct sequencing are listed in Online Supplementary Table S2.

Outcome analyses

Comparisons between groups were performed with the Fisher exact and Mann-Whitney tests for categorical and continuous variables respectively. Corticosteroid sensitivity was defined as clearance of peripheral blood circulating blasts (<1 × 10/L) following steroid prophase treatment. Complete remission was defined as clearance of bone marrow blasts (<5%) following induction treatment. Overall survival was calculated from the date of inclusion in the trial to the last follow-up date, censoring patients alive at that date. Event-free survival was calculated from date of inclusion in the trial to the date of induction failure, relapse, or death, censoring patients alive in first complete remission without relapse at the last follow-up date. Cumulative incidence of relapse was calculated in patients who attained complete remission, from the date of achieving the complete remission to the date of relapse, with death in first complete remission being considered as a competing event. Univariate and bivariate analyses assessing the impact of DNMT3A mutations and age were performed with a Cox model. Variables that were significantly associated with outcome in univariate analysis were considered as covariates in multivariate Cox models. The proportional-hazards assumption was checked before conducting multivariate analyses. Statistical analyses were performed with STATA software (STATA 12.0 Corporation, College Station, TX, USA). All P values were two-sided, with P<0.05 denoting statistical significance.

Results

Analysis of DNMT3A genotype in patients with T-cell acute lymphoblastic leukemia in the GRAALL studies

We performed targeted NGS of a panel of genes, including DNMT3A, which have been described to be recurrently mutated in T-ALL. This panel included all exons of DNMT3A, thereby providing a comprehensive picture of the spectrum of alterations across this gene in T-ALL. Diagnostic DNA was available for 198 patients treated during the GRAALL-2003 and -2005 studies. A partial analysis of a subgroup of this cohort has been reported previously.24 We detected 21 DNMT3A mutations in 18 patients (9.1%). Most alterations occurred in regions coding for defined protein functional domains, including six mutations at the R882 hotspot1 (Figure 1A). Further details of patient-specific alterations are shown in Online Supplementary Table S3. Of note, the vast majority of detected mutations are predicted to be significantly damaging to protein function.

Figure 1.DNMT3A mutations in T-cell acute lymphoblastic leukemia. (A) Schematic representation of the 21 mutations detected in this study. Further patient-specific details are provided in Online Supplementary Table S3. (B) Comparison of the mutational genotypes of DNMT3A altered (n=18) and DNMT3A wild-type (n=180) T-cell acute lymphoblastic leukemia. Percentage frequencies in each group are depicted. Functional categories are listed in bold.

In keeping with previous reports of DNMT3A-mutated T-ALL which cited high rates of either compound heterozygosity or homozygosity,119 a significant proportion of cases (8/18) had either two separate alterations, or high variant allele frequencies that were suggestive of either homozygous mutation, concomitant deletion of the wild-type (WT) allele or copy-neutral loss of heterozygosity. Comparative genomic hybridization analyses were available for 85 of the cases in this study, including 6/18 patients with DNMT3A mutations. We detected only two deletions of the DNMT3A locus, which in each case were associated with concomitant DNMT3A mutation and elevated variant allele frequencies (cases 11 and 12 in Online Supplementary Table S3).

The prevalence of other mutations detected by NGS is shown in Figure 1B. DNMT3A-altered cases had an increased frequency of alterations in other genes included in the NGS panel, compared with the rest of the cohort (88.9% DNMT3A mutated vs. 64.4% DNMT3A WT, P=0.036). There were no statistically significant differences in the prevalence of mutations in any specific functional gene category, namely factors involved in cytokine receptor and RAS signaling (61.1% DNMT3A mutated vs. 41.7% DNMT3A WT, P=0.113), hematopoiesis (38.5% DNMT3A mutated vs. 12.8% DNMT3A WT, P=0.082) and chemical modification of histones (50.0% DNMT3A mutated vs. 30.0% DNMT3A WT, P=0.082). However, we did observe significant co-occurrence of DNMT3A alterations and IDH2 mutations (27.8% DNMT3A mutated vs. 2.2% DNMT3A WT, P<0.001). This association has been described previously in both AML25 and myelodysplastic syndromes.26

Evidence of possible clonal hematopoiesis in DNMT3A-mutated T-cell acute lymphoblastic leukemia

DNMT3A is the most commonly altered gene in age-related clonal hematopoiesis,2927 and DNMT3A mutations have been detected in non-malignant cells in AML3130 and peripheral T-cell lymphoma.14 We therefore tested whether DNMT3A mutations were present in non-leukemic hematopoietic cells in T-ALL patients.

DNA from remission bone marrow was available for only three of the 18 patients with DNMT3A mutations. While two of these samples had a DNMT3A WT genotype (Online Supplementary Figure S1), one interesting case had evidence of DNMT3A alterations in non-leukemic bone marrow cells. At diagnosis, this patient (case 6 in Online Supplementary Table S3) had mutations in exons 14 and 15 of DNMT3A, a NOTCH1 PEST domain insertion, and an NRAS G12D substitution. Sequencing of remission DNA revealed mutation of DNMT3A exon 14 in non-leukemic cells, while NOTCH1, NRAS and DNMT3A exon 15 all presented wild-type genotypes (Figure 2A). We confirmed that the exon 14 mutation has never been reported as a polymorphic variant, while SIFT analysis (http://sift.jcvi.org/) predicted this M548T substitution to be highly deleterious to protein function, with a SIFT score of 0. These results suggest that this T-ALL may have developed on a background of DNMT3A-mutated clonal hematopoiesis, and that the other genetic alterations, including the second DNMT3A mutation, were acquired at leukemic transformation.

Figure 2.Evidence of DNMT3A mutations in non-leukemic DNA. (A) Direct sequencing of DNMT3A exons 14 and 15, NOTCH1 and NRAS in diagnostic (left panels) and remission (right panels) samples. (B) Mutational assessment of DNA extracted from leukemic and non-leukemic fractions of samples from patients with T-cell acute lymphoblastic leukemia. Sequencing results of DNMT3A and NOTCH1 in leukemic (left panels) and non-leukemic (right panels) DNA are shown. Cases are numbered according to the listing in Online Supplementary Table S3.

In order to extend this analysis of non-leukemic DNMT3A mutation, we performed immunophenotypic sorting of two further diagnostic bone marrow samples, and extracted DNA from both the leukemic and the minor residual non-leukemic fractions. We detected a mutation in non-leukemic DNA in one patient (case 4 in Online Supplementary Table S3). Again, we confirmed that this mutation has not been reported as a polymorphism, and that the resultant P385L substitution is predicted to damage protein function, with a SIFT score of 0.02. Similar to the case with mutated remission DNA, this sample was negative for two NOTCH1 alterations detected at T-ALL diagnosis, confirming the specificity of DNMT3A mutation persistence (Figure 2B). The other tested non-leukemic DNA had a DNMT3A WT genotype (Online Supplementary Figure S2), giving an overall rate of non-leukemic DNMT3A mutant positivity of 2/5 samples from the GRAALL-2003 and -2005 studies. We also tested a further three T-ALL cases not included in this cohort, but found no evidence of non-leukemic DNMT3A mutation. This gives an overall incidence of possible clonal hematopoiesis in 2/8 T-ALL samples assessed in our laboratory. It was unfortunately not possible to obtain non-hematopoietic tissue from either of these patients, in order to exclude that these alterations were not constitutional, and to confirm definitively that these results reflect the persistence of a DNMT3A-mutated clonal hematopoietic population in these cases.

DNMT3A mutations are associated with older age and treatment resistance

A clinico-biological comparison of cases with and without DNMT3A mutations is shown in Table 1. In keeping with previous reports,1211 patients with mutations were considerably older than the rest of the T-ALL cohort (median age 43.9 years mutated vs. 29.4 years non-mutated, P<0.001). In addition, DNMT3A-mutated leukemias were more likely to have an immature T-receptor genotype32 (53.3% mutated vs. 24.4% non-mutated, P=0.016), although this did not correspond to a significantly higher incidence of an ETP-ALL immunophenotype33 (35.7% mutated vs. 20.3% non-mutated, P=0.184).

Table 1.Characteristics and outcome of the patients according to DNMT3A genotype.

DNMT3A mutation was notably associated with poor initial treatment response. We observed trends towards early corticosteroid resistance (66.7% mutated vs. 43.3% non-mutated, P=0.081) and induction failure (13.3% vs. 2.9%, P=0.096), and patients with DNMT3A mutations had significantly higher rates of death during induction (16.7% vs. 2.8%, P=0.027), and lower attainment of complete remission (72.2% mutated vs. 94.4% non-mutated, P=0.006). As only four patients with mutations were evaluated for minimal residual disease, we could not verify that molecular remission was similarly compromised.

We found that the type of DNMT3A mutation did not significantly correlate with any individual clinico-biological parameter, suggesting that the alterations detected in this study are likely to have broadly similar biological consequences.

DNMT3A mutation correlates with poor outcome in T-cell acute lymphoblastic leukemia

The median follow-up of the cohort was 5.5 years. Prognostic analyses revealed that DNMT3A mutation was associated with an increased 5-year cumulative incidence of relapse (53.9% mutated vs. 28.7% non-mutated, P=0.037) (Figure 3A) and with 5-year event-free survival [27.8% mutated vs. 61.0% non-mutated; hazard ratio (HR) 3.22, (95% confidence interval (95% CI): 1.81-5.72, P<0.001] (Figure 3B). Patients with DNMT3A mutations also had a markedly inferior 5-year overall survival (38.8% mutated vs. 68.7% non-mutated, HR 2.91, 95% CI: 1.56-5.43, P=0.001) (Figure 3C).

Figure 3.DNMT3A mutation correlates with poor outcome in T-cell acute lymphoblastic leukemia. Comparisons of outcomes for patients with (n=18) and without (n=180) DNMT3A mutations are shown for: (A) cumulative incidence of relapse; (B) event-free survival; and (C) overall survival. The 5-year results were as follows: cumulative incidence of relapse 53.9% mutated vs. 28.7% non-mutated; event-free survival 27.8% mutated vs. 61.0% non-mutated; overall survival 38.8% mutated vs. 68.7% non-mutated. P values are indicated.

The poor prognosis of DNMT3A-mutated T-cell acute lymphoblastic leukemia is age-independent

Our and others’ data1211 have shown that the incidence of DNMT3A mutation in T-ALL increases with age, but previous reports have not documented whether this factor contributes to prognosis. As older patients treated during the GRAALL studies had worse outcomes due to impaired tolerance of intensive chemotherapy,34 we considered it critical to determine to what extent age was a confounding prognostic variable.

We therefore performed bivariate analyses of the effects of DNMT3A mutations and age across a series of outcome measures. These results are shown in Online Supplementary Table S4. In each case, DNMT3A genotype was still associated with significantly increased cumulative incidence of relapse (HR 2.80, 95% CI: 1.12-6.97, P=0.034), and shorter event-free survival (HR 2.62, 95% CI: 1.45–5.06, P=0.004) and overall survival (HR 2.05, 95% CI: 1.02-4.12, P=0.043).

Since DNMT3A alterations were almost exclusively found in patients >40 years (16/18 cases), we also performed survival analyses that were restricted to the >40-year old subgroup, which constituted a quarter of the total cohort of patients (50/198, 25.3%). Consistent with the results of the bivariate analyses, DNMT3A mutation was associated with significantly worse 5-year cumulative incidence of relapse (58.3% mutated vs. 21.7% non-mutated, HR 3.90, 95% CI: 1.30-11.68, P=0.015) (Figure 4A), 5-year event-free survival (25.0% mutated vs. 56.7% non-mutated, HR 2.95, 95% CI: 1.37-6.32, P=0.005) (Figure 4B), and 5-year overall survival (37.5% mutated vs. 62.1% non-mutated, HR 2.35, 95% CI: 1.05-5.26, P=0.038) (Figure 4C).

Figure 4.DNMT3A genotype predicts outcome in the age group of patients at risk of mutation. Comparisons of outcomes for patients with (n=16) and without (n=34) mutations in patients >40 years are shown for: (A) cumulative incidence of relapse; (B) event-free survival; and (C) overall survival. The 5-year results were as follows: cumulative incidence of relapse 58.3% mutated vs. 21.7% non-mutated; event-free survival, 25.0% mutated vs. 56.7% non-mutated; overall survival 37.5% mutated vs. 62.1% non-mutated. P values are indicated.

Finally, we carried out multivariate outcome analyses in the whole cohort using the risk factors that were used to stratify treatment during the GRAALL-2003 and -2005 studies, and which were found to significantly predict prognosis in the univariate analyses. Among age, log(WBC), corticosteroid sensitivity, early chemosensitivity, and DNMT3A genotype, only DNMT3A genotype was associated with cumulative incidence of relapse in univariate analysis (data not shown). As shown in Tables 2 and 3, age, log(WBC), corticosteroid resistance along with DNMT3A genotype were significantly associated with a poor event-free survival and overall survival. In multivariate analysis adjusting for these covariates, DNMT3A mutation was still significantly associated with shorter event-free survival (HR 2.33, 95% CI: 1.06–4.04, P=0.02) (Table 2), although not with overall survival (HR 1.66, 95% CI: 0.82–3.37, P=0.16) (Table 3).

Table 2.Prognostic impact of DNMT3A genotype on event-free survival.

Table 3.Prognostic impact of DNMT3A genotype on overall survival.

Taken together, these results provide strong evidence that DNMT3A mutation, while mostly observed in older cases, predicts a poor prognosis that is not related to the patient’s age.

Discussion

To our knowledge, this is the most extensive study of DNMT3A-mutated T-ALL yet reported. Our targeted NGS approach allowed comprehensive assessment of genotype across the entire DNMT3A locus, along with the prevalence of co-occurring genetic alterations. Our data additionally benefit from the analysis of a large cohort of patients who were uniformly treated as part of the GRAALL-2003 and -2005 studies, thereby allowing rigorous outcome comparisons between mutated and wild-type cases.

Some of our results were expected, and the findings that DNMT3A mutations are more commonly present in older patients and genotypically immature leukemias are consistent with previously published data.13119 We did not, however, observe increased rates of ETP-ALL immunophenotype, as might have been predicted. We did not detect a clear association with any other genetically-defined subgroup, and there was no link to increased HOXA expression, which we have previously shown to predict outcome in immature T-ALL.35

The detection of DNMT3A alterations in non-leukemic bone marrow suggests that some of these cases of T-ALL might have arisen from DNMT3A-mutated clonal hematopoiesis. While pre-leukemic NOTCH1 mutations have been detected in neonatal blood spot samples of pediatric patients with T-ALL,36 to our knowledge our data provide the first potential evidence of age-related clonal hematopoiesis in T-ALL. As it was not possible to obtain non-hematopoietic tissue from either of the patients with this finding, we cannot definitively exclude that these alterations are constitutional, or might even represent an inherited cancer predisposition. Further work is necessary to investigate the incidence of clonal hematopoiesis linked to alterations in DNMT3A and other genes in T-ALL.

Non-leukemic DNMT3A mutations have been seen in AML,3130 and it has been postulated that DNMT3A-altered immature T-ALL might arise from malignant transformation of a multipotent myeloid/lymphoid progenitor cell.12 In keeping with this, one of the cases with a non-leukemic DNMT3A mutation in this study had expression of myeloid cell surface markers as part of an ETP-ALL phenotype, while full immunophenotypic assessment was unfortunately not possible for the other patient. The factors that may dictate the acute leukemic phenotype in clonally mutated cases remain to be clarified. For example, this might be influenced by the differentiation capacity of the cell in which the initial DNMT3A mutation occurs. In addition, it is tempting to speculate that the acquisition of specific cooperative mutations, such as the NOTCH1 mutations observed in these T-ALL cases, might act as lineage determinants.

Outcome analyses revealed that DNMT3A mutation correlated with poor prognosis independently of the patients’ age in bivariate analyses. Multivariable analyses using parameters that were used to stratify treatment in the GRAALL-2003 and -2005 studies showed that DNMT3A genotype independently predicted both event-free survival and cumulative incidence of relapse. DNMT3A mutation status also independently predicted event-free survival and overall survival in bivariable analyses that incorporated our recently described oncogenetic risk classifier37 (Online Supplementary Table S5). These results suggest that DNMT3A mutation is directly linked to aggressive T-ALL biology. As DNMT3A-altered T-ALL had higher mutation rates in other genes included in our targeted sequencing panel, it is also possible that increased genotype complexity may contribute to the more aggressive phenotype in these leukemias. This issue may be clarified by more comprehensive genomic assessment in future studies.

The high rates of treatment failure observed in this study suggest that therapeutic intervention is warranted for DNMT3A-mutated cases, and that treatment intensification should be considered for the infrequent younger patients with mutations. Indeed, we have previously documented a benefit from allogeneic stem cell transplantation in first complete remission for ETP-ALL,24 which similarly exhibits high rates of intrinsically treatment-resistant disease. As only three of the 18 DNMT3A-mutated patients in this study underwent allogeneic stem cell transplantation (data not shown), we are unable to estimate the potential benefit of such treatment in this setting. We recently reported that treatment-related toxicity in the GRAALL-2005 study increased in proportion to the patients’ age,38 and further therapy intensification in elderly patients must therefore be considered of questionable benefit. The upper age of this study cohort was 60 years, but it is likely that the rate of mutations in older patients who do not tolerate such intensive chemotherapy is higher. Data reported for patients with AML suggest that DNMT3A mutation confers increased sensitivity to hypomethylating agents,39 providing a rationale for evaluation of these drugs in DNMT3A-mutated T-ALL. In the longer term, it is to be hoped that investigation of the molecular mechanisms by which DNMT3A mutation alters T-ALL biology will lead to better treatments and improved outcomes for these high-risk cases.

Acknowledgments

This manuscript was written on behalf of the Group for Research on Adult Acute Lymphoblastic Leukemia (GRAALL), which includes the former France-Belgium Group for Lymphoblastic Acute Leukemia in Adults (LALA), the French Western-Eastern Group for Lymphoblastic Acute Leukemia (GOELAL), and the Swiss Group for Clinical Cancer Research (SAKK). The authors would like to thank all participants in the GRAALL-2003 and GRAALL-2005 study groups for collection and provision of data and samples, and V. Lheritier for collection of clinical data. The GRAALL-2003 study was sponsored by the Hôpitaux de Toulouse, and the GRAALL-2005 study by the Assistance Publique-Hôpitaux de Paris. The SAKK was supported by the Swiss State Secretariat for Education, Research and Innovation (SERI). JB was supported by a Kay Kendall Leukaemia Fund Intermediate Research Fellowship and by the National Children’s Research Centre, Children’s Health Ireland at Crumlin, Dublin, Ireland. The Necker Laboratory is supported by the Association Laurette Fugain, La Ligue contre le Cancer and the INCa CARAMELE Translational Research and PhD programs.

Footnotes

• Check the online version for the most updated information on this article, online supplements, and information on authorship & disclosures: www.haematologica.org/content/104/8/1617

• Received May 24, 2018.
• Accepted January 10, 2019.